Parametric generation of training data for ML-based OSV motion prediction

Abstract

Maintenance operations are crucial to ensure a steady energy supply provided by offshore wind parks. These operations involve the transfer of workers from one floating structure to another, such as from a hospitalization platform to an Offshore Supply Vessel (OSV) and vice versa. The primary motivation for this research is to accelerate the early-stage design process by using machine learning (ML), and for this purpose, the OSVs are chosen as a reference structure. To effectively train ML models, comprehensive and accurate data is essential. This research focuses on generating this data using parametric modeling and potential theory, exemplified by the design and motion analysis of OSVs.

Traditionally, methods like potential theory and Response Amplitude Operators (RAOs) are used for motion prediction. While useful for studying frequency domain behavior, these methods are time-consuming and do not account for non-linear wave effects that significantly impact real vessels. But the purpose of this research is to use already established parametric modeling tools, such as CAESES, and hydrodynamic analysis software, like WAMIT, to create a wide range of dataset for ML training. WAMIT is a also boundary element (BEM) solver, and provides satisfactory results. As an outcome, this study generated hundreds of parametric OSV models, providing a robust data foundation for developing AI models capable of accurate and efficient motion prediction. While data accuracy can be refined in future stages, the current focus is on setting up a robust analytical framework.

The next phase of this study can potentially involve validating the AI model by comparing its predicted RAOs with those generated by WAMIT. Successful validation will demonstrate the feasibility of using AI for efficient and accurate motion prediction, thereby reducing the design time for OSVs and potentially other offshore structures.